Class 12 Flashcards
computer vision
perceptual channel that accepts a stimulus and reports some representation of the world
passive sensing
type of vision that most agents use where they don’t need to send out light to see
feature
number obtained by applying simple computations to an image
feature extraction
process of applying direct computation to sensor responses
object model
precise geometric model produced by computer aided design systems
rendering model
describes the physical, geometric, and statistical processes that produce the stimulus from the word
reconstruction
when an agent builds a model of the world from an image or images
recognition
term for drawing distinctions among objects based on visual and other information
pinhole camera
simplest way to form a focused image of a stationary object
aperture
opening or “pinhole”
motion blur
object is moving and the time window is large enough
focal length
distance between lens and the image sensor
focal plane
plane in the environment where the focus is sharpest
scaled orthographic projection
used to handle geometric effect of perspective imaging that aren’t always pronounced
diffuse reflection
light is scattered evenly in all directions – how most surfaces reflect light
specular reflection
causes incoming light to leave a surface in a lobe of directions that is determined by the direction the light arrived from
specularities
small bright patches found on rougher surfaces
lamberts cosine law
brightness of a diffuse patch
interreflections
light reflected from other surfaces which illuminates shadowed patches
principle of trichromacy
humans of 3 types of color sensitive receptors
edge
straight line or curve in image plane across which there is a significant change in image brightness
texture
pattern on a surface that can be sensed visually
texels
rough model of texture that is a repetitive pattern of elements
optical flow
whenever there is relative movement between the camera and 1 or more objects in the scene resulting in apparent motion
segmentation
process of breaking an image into groups of similar pixels
regions
generalized groupings of segments once edges have been determined – aka superpixels
convolutional neural networks
excellent image classifier, pooling layers reduce spatial dimensions of input
regional proposal network
network that finds regions with objects
region of interest
a box with a good enough objectness score
roi pooling
process of sampling pixels to extract features
bounding box regression
step that trims the window down to a proper bounding box
disparity
the shifting of one view to another view
tagging systems
tag images with relevant words
depth map
an array giving the depth of each pixel
style transfer
involves taking 2 images with content and style and transferring the content of one onto the style of the other
